Glycogen-dependent demixing of frog egg cytoplasm at increased crowding

Crowding increases the tendency of macromolecules to aggregate and phase separate, and high crowding can induce glass-like states of cytoplasm. To explore the effect of crowding in a well-characterized model cytoplasm we developed methods to selectively concentrate components larger than 25 kDa from Xenopus egg extracts. When crowding was increased 1.4x, the egg cytoplasm demixed into two liquid phases of approximately equal volume. One of the phases was highly enriched in glycogen while the other had a higher protein concentration. Glycogen hydrolysis blocked or reversed demixing. Quantitative proteomics showed that the glycogen phase was enriched in proteins that bind glycogen, participate in carbohydrate metabolism, or are in complexes with especially high native molecular weight. The glycogen phase was depleted of ribosomes, ER and mitochondria. These results inform on the physical nature of a glycogen-rich cytoplasm and suggest a role of demixing in the localization of glycogen particles in tissue cells.


Introduction
The concentration of macromolecules in cytoplasm is thought to reflect a balance between competing 12 evolutionary pressures. Broadly speaking, increased concentrations tend to speed up biochemistry, but 13 very high concentrations can cause excessive crowding, leading to deleterious interactions and potentially 14 freezing of biochemistry that depends on dissociation reactions of macromolecules. In more precise 15 terms, reaction-limited reaction rates increase with crowding due to increased re-collision frequency of 16 reactants (Kim and Yethiraj, 2009). At the same time, diffusion-limited reaction rates increase with 17 concentration but decrease with viscosity, and viscosity increases with concentration (Dill et al., 2011;van 18 den Berg et al., 2017). Crowding tends to promote macromolecular assembly reactions, including linear 19 polymerization and phase-separated condensate assembly (Andre and Spruijt, 2020). Increasing crowding 20 may contribute to a liquid-to-glass transition in bacterial and yeast cytoplasm under energy stress (Joyner 21 et al., 2016;Munder et al., 2016;Parry et al., 2014) or mechanical compression (Okumus et al., 2016). 22 Whether this glass transition is a universal response to increased crowding is not known, though local 23 dehydration was shown to decrease the mobility of macromolecules in human cells in tissue culture 24 (Charras et al., 2009). 25

26
The concept of crowding is related to those of excluded volume effects (Rivas and Minton, 2016) and 27 colloid osmotic pressure (Mitchison, 2019). Importantly, the ability of a molecule to exert crowding 28 effects, and the response of molecules to crowding, depend on their size. Cytoplasm is polydisperse and 29 soluble biomolecules cover orders of magnitude in radius, spanning from water molecules to ribosomes 30 and other large complexes. Increased crowding tends to preferentially affect larger molecules because 31 they start to physically interact before smaller molecules (Hwang et al., 2016). This reduces the mobility 32 of large molecules, such as ribosomes, at concentrations which still allow smaller molecules to move in 33 the crevices between them (Delarue et al., 2018). One might expect increased crowding to induce 34 demixing of the largest components while leaving smaller components free to diffuse; however, the effect 35 of crowding on cytoplasm is complicated by binding reactions, which may cause smaller molecules to 36 demix along with larger ones to which they bind. 37

38
The storage polysaccharide glycogen is one of the most abundant macromolecules in many animal cells, 39 on the order of 10% (w/w) in fed liver tissue (Dowler and Mottram, 1918;Prats et al., 2018); however, it 40 is often forgotten in discussions of the structure and dynamics of cytoplasm. Glycogen is synthesized as a 41 particle that consists of a highly branched, covalent polymer of glucose, usually initiated by polymerization 42 from the surface of the nucleation protein glycogenin (Prats et al., 2018). Glycogen particles can be of 43 different sizes, with a diameter of ~20 nm typical (Ioan et al., 1999a, b). Glycogen particles are highly 44 soluble in water, though they can form higher order assemblies (Nawaz et al., 2021;Prats et al., 2018). 45 The cell biology and biophysics of glycogen have been little studied in recent years, though a recent report 46 of the glycogen-binding proteome is relevant to this work (Stapleton et al., 2013;Stapleton et al., 2010). 47

48
Xenopus egg extracts provide a well-characterized model cytoplasm. They are prepared from eggs by 49 centrifugal crushing with minimal dilution and maintain many of the biochemical and biophysical 50 properties of native egg cytoplasm. Egg extracts contain ~80 mg/mL protein and ~80 mg/mL glycogen. 51 The glycogen provides an energy store for the developing embryo and may also provide a crowding 52 function that promotes assembly of nuclei (Hartl et al., 1994) and mitotic spindle poles (Groen et al., 53 2011). We set out to investigate the effect of changes in crowding on cytoskeleton dynamics in egg extract, 54 but when we increased crowding by 1.4x or higher, we observed bulk demixing into two phases, which 55 was unexpected. Here, we report that this demixing depends on glycogen and generates one phase that 56 is highly glycogen-enriched. These observations probably do not model any normal egg biology, but they 57 are informative concerning the physical properties of glycogen and its influence on those of the cytoplasm. 58

Results
Crowded Xenopus egg extracts exhibit liquid-liquid demixing 59 We developed two methods to selectively increase macromolecular crowding of Xenopus egg extracts 60 while minimally perturbing ionic strength, pH and metabolite concentrations. Both methods gave similar 61 results. In the first method, dry Sephadex G-25 gel filtration resin was added directly to extract and then 62 removed a few minutes later by centrifugal filtration. As the resin swells, it absorbs water and small 63 molecules, but macromolecules greater than 25 kDa are excluded and therefore become more crowded 64 ( Fig 1A). This method was convenient for small volumes of extract. The crowding factor depended on the 65 amount of dry resin added per volume extract (Figs 1B, S1). In the second method, a 30 kDa MWCO 66 centrifugal filter unit was used to concentrate macromolecules greater than 30 kDa. This method was 67 convenient for larger volumes. The crowding factor was measured by adding a macromolecular 68 fluorescent probe, such as Streptavidin (53 kDa) fused to Alexa Fluor 647 (Fig 1C), then comparing the 69 fluorescence intensity of a specimen of fixed depth before and after crowding by fluorescence microscopy 70 with a low magnification objective ( Fig S1). 71 72 When crowding was increased to 1.2x, the extract remained similar in appearance to uncrowded extracts, 73 as observed in brightfield and fluorescence images (Fig 1B,C). At 1.4x-1.5x, the extract underwent 74 spontaneous demixing over a few minutes at both 0 °C and 20 °C. This unexpected phenomenon was 75 further characterized. At 1.9x the extract appeared to precipitate, and this regime was not examined 76 further. The phases formed throughout the sample and remained co-mingled in the tube. They had 77 different densities and could be separated in bulk by centrifugation at 20000 rcf for 20 min. The volumes 78 of the two phases after centrifugation were similar. The denser phase had a higher index of refraction of 79 1.39, and the less dense phase had a lower index of refraction of 1.38. This refractive index difference 80 made demixing easy to follow by phase contrast or DIC microscopy. Both phases exhibited all the 81 hallmarks of liquids, including deformation under flow, splitting, fusion, and rounding towards a spherical 82 shape driven by surface tension (Hyman et al., 2014) (Fig 1D, Video S1). 83 The macromolecular fraction greater than 25 kDa of Xenopus egg extracts was crowded by adding dry Sephadex G-25 resin, which selectively imbibes water and small molecules as it swells. (B) Brightfield images. Above 1.4x crowding, the extract remained liquid-like and exhibited patches with different light scattering properties. Above 1.9x crowding, precipitation occurred. (C) The crowding factor was estimated by quantification of fluorescence of a protein probe greater than 25 kDa added before crowding, in this case Streptavidin (53 kDa) fused to Alexa Fluor 647. The Streptavidin-A647 partitioned into one of the phases. Crowding factors estimated from lower magnification images in Fig S1. (D) Time lapse differential interference contrast (DIC) images of a 1.4x crowded extract confined between coverslips. The phases exhibited hallmarks of liquids, including deformation under flow, splitting, fusion, and rounding by surface tension. See Video S1.

Role of glycogen in demixing 84
The higher refractive index of the denser phase suggested non-equal distribution of glycogen, which has 85 a higher density and refractive index than protein. We measured the glycogen concentration in 86 unperturbed extract and the two phases using an assay that digested it to glucose for colorimetric 87 quantification. Glycogen was highly enriched in the denser phase (Fig 2A). The glycogen concentration 88 was 80 mg/mL in uncrowded crude extracts, 20 mg/mL in the less dense phase, and 250 mg/mL in the 89 denser phase (Fig 2A) (Methods). Total protein was slightly enriched in the less dense phase (Fig 2B). 90

91
To test for a role of glycogen in demixing, we hydrolyzed it using the enzyme amyloglucosidase (AG) 92 (Methods). Glycogen digestion by AG blocked demixing when added before crowding (not shown) and 93 reversed it when added after crowding, so demixing depended on glycogen ( Fig 2C). 94 Figure 2. Role of glycogen in demixing. (A) Glycogen highly partitioned between the phases, as measured by a colorimetric assay. The concentration of glycogen in uncrowded extract was 80 mg/mL. In crowded extracts, the glycogen concentration was 20 mg/mL in the less dense phase and 250 mg/mL in the denser phase. (B) The less dense, glycogen-depleted phase had a higher protein concentration than the denser, glycogen-enriched phase. (C) DIC images. Top row: Control crowded extract remained demixed. Bottom row: Addition of amyloglucosidase (AG) after demixing caused the phases to dissolve and the system to return to a single phase.

Ultrastructure of glycogen-enriched (G) and -depleted (R) phases 95
Thin-section electron microscopy with conventional heavy metal staining was used to probe the 96 ultrastructure of the phases. Uncrowded crude extracts appeared mottled with uniformly distributed 97 mitochondria and ER (Fig 3A). Crowded extracts exhibited at least two major phases as in optical 98 micrographs ( Fig 3B). One of the phases had higher electron density than the other (Fig 3B). To identify 99 the phases in electron micrographs, the phases were isolated in bulk by centrifugation then imaged 100 separately (Methods). The glycogen-depleted phase (Fig 3C) had higher electron density than the 101 glycogen-enriched phase (Fig 3D). Mitochondria concentrated at the interface between the phases (Fig  102   3B,E) and were also present within the glycogen-depleted phase (Fig 3C). The higher electron density, 103 glycogen-depleted phase was textured with structures ~25 nm in diameter, which we interpret as 104 ribosomes (Fig 3E''). Hereafter, we refer to the glycogen-depleted phase as "R" for ribosomes and the 105 glycogen-enriched phase as "G" for glycogen. 106 The glycogen-depleted and glycogen-enriched phases were isolated from one another in bulk by centrifugation and imaged separately. The glycogen-depleted phase was higher electron density than the glycogen-enriched phase. (E) Mitochondria often localized along the interface between the phases or in the higher contrast phase. E' is a zoom of the box in panel E, and E'' is a zoom of the box in panel E'. The higher contrast phase had a granular appearance with features ~25 nm in diameter, which we interpret as ribosomes.

Fluorescent probes partitioned between G and R phases 107
Fluorescence microscopy provided a convenient method to observe the two phases and estimate the 108 partition coefficient of macromolecules. Fluorescent probes such as EB1-mApple (57 kDa) and 109 Streptavidin (53 kDa) labeled with Alexa Fluor 647 were added to crowded extracts. Then, to estimate 110 partitioning of each probe between the phases, the G and R phases were isolated from one another in 111 bulk by centrifugation (Methods). We could thus estimate the partition coefficients of the fluorescent 112 probes (Fig 4A,B), as well as identify the mixed phases using the fluorescent probes (Fig 4C). Partition 113 coefficients in mixed phases were similar to those in bulk (Fig 4D). Most probes partitioned preferentially 114 into the R phase, including EB1-GFP (57 kDa), Fab fragment antibody (50 kDa) labeled with Alexa Fluor 115 647, and 70 kDa dextran-Alexa Fluor 488 (Fig 4E-G, I-J). Glycogen phosphorylase A (PYGL, 188 kDa as 116 dimer), a glycogen-binding protein, labeled with Pacific Blue partitioned preferentially into the G phase 117 ( Fig 4K). Mitochondria imaged by NADH autofluorescence localized along the interface between phases 118 ( Fig 4H), as seen by electron micrographs (Fig 3B,E). 119

Protein partitioning depends on glycogen binding and native molecular weight (MW) 120
To quantify the proteomes of the G and R phases, we performed multiplexed mass spectrometry analysis 121 using the MultiNotch MS3 method McAlister et al., 2014;Sonnett et al., 2018). For 122 this analysis, we compared two methods for crowding the extract, using either Sephadex G-25 resin or 30 123 kDa MWCO centrifugal filter units. Results were similar for the two methods ( Fig S2, Table S1). Fig 5  124 reports measurements averaged across the two methods and several repeats. 125 126 Known glycogen-binding proteins partitioned into the G phase, with log base 2 partition coefficients of 1-127 3 ( Fig 5A). These values approach the partition coefficient of glycogen itself (Fig 2A). Ribosomal subunits, 128 ER and mitochondrial proteins selectively partitioned into the R phase (Fig 5A), consistent with the 129 electron micrographs (Fig 3C,E). Enzymes involved in carbohydrate metabolism also partitioned into the 130 G phase, though with lower partition coefficients than proteins known to bind glycogen ( Fig 5A). Several 131 of the highest native MW protein complexes other than ribosomes partitioned into the G phase, such as 132 major vault protein (MVP, 13 MDa), ferritin (FTH1, 450 kDa), chaperonin-containing T-complex (CCT, 960 133 kDa), and the 26S proteasome (PSMA, 2 MDa) (Fig 5A). We then plotted log base 2 partition coefficients 134 with respect to native MW (Fig 5B), based on previous estimates of native MW with an upper bound of 135 256 kDa (Wühr et al., 2015). Proteins with native MW smaller than 100 kDa had a slight preference for 136 the G phase, with average log base 2 partition coefficient 0.4 ± 0.8 (Fig 5C). In contrast, most proteins with 137 native MW greater than 256 kDa partitioned into the R phase, with average log base 2 partition coefficient 138 -1.3 ± 1.6 ( Fig 5C). 139

Discussion
When macromolecular crowding of Xenopus egg extracts was increased 1.4x over control, we observed 140 glycogen-dependent demixing into liquid G and R phases that were enriched in glycogen and ribosomes, 141 respectively. This was unexpected, since reports from other systems led us to expect a transition to a 142 glass-like state. We suspect liquid-liquid demixing is promoted by the high glycogen concentration in egg 143 cytoplasm, which is comparable to that in hepatocytes from fed liver. 144 145 Mass spec analysis suggested both binding and native MW contributed to partitioning of proteins 146 between the phases. In terms of binding, glycogen-binding proteins partitioned preferentially into the G 147 phase with an average partition coefficient similar to that of glycogen itself (Fig 5A). Enzymes involved in 148 carbohydrate metabolism also partitioned into the G phase ( Fig 5A). These spanned a range of native MWs 149 from 33 to 241 kDa, with no apparent correlation between partition coefficient and native MW (Table S1). 150 Some of these have been shown to bind glycogen in an adipocyte glycogen proteome (Stapleton et al., 151 2013;Stapleton et al., 2010). Association between glycolytic enzymes has also been reported, but its 152 functional significance remains unclear (Schmitt and An, 2017). Consistent with a role for native MW, the 153 G phase enriched several especially large protein complexes not known to bind glycogen (Stapleton et al., 154 2013;Stapleton et al., 2010) (Fig 5A). Relative contributions of binding and entropic considerations may 155 be considered within excluded volume theory (Rivas and Minton, 2016). 156 157 Frog eggs evolved in a freshwater environment and Xenopus eggs are not known to exhibit desiccation 158 resistance. Thus, the demixing we observed is unlikely to be of direct physiological relevance; however, it 159 may provide clues to glycogen organization in tissue cells. Glycogen particles often appear as aggregates 160 in multiple animal tissues (Coimbra and Leblond, 1966;Galavazi, 1971;Porter and Bruni, 1959;Revel, 161 1964;Revel et al., 1960;Sheldon et al., 1962) and chloroplasts (Crumpton-Taylor et al., 2012;Kasperbauer 162 and Hamilton, 1984). Fawcett (1981) summarized extensive EM studies as showing that "glycogen is 163 seldom uniformly distributed in the cytoplasm but tends to accumulate in dense regional deposits." Our 164 results suggest that physical demixing may contribute to high local concentrations of glycogen, though we 165 cannot rule out other mechanisms including binding interactions between glycogen particles and local 166 concentration of biosynthetic enzymes. 167

168
Our observations are also relevant to Xenopus egg extract technical considerations. Concentration of 169 extract using 100 kDa filtration units was shown to increase the stability of extracts to freeze-thaw cycles 170 (Takagi and Shimamoto, 2017). Those authors used ~1.2x crowding, which is just below the concentration 171 factor needed for demixing. Exploration in the crowded but still mixed regime may facilitate study of the 172 effect of crowding on biochemical processes. Recent work examined how crowding affects microtubule 173 polymerization using osmotic perturbation of fission yeast (Molines et al., 2020). It will be interesting to 174 ask similar questions in cytoplasmic extracts. 175

Preparation of Xenopus egg extracts 176
Xenopus egg extracts were prepared as described previously (Field et al., 2017). Most experiments used 177 extracts prepared with Cytochalasin D to prevent F-actin polymerization, in which case 100 µg/mL 178 Cytochalasin D was added before the crushing spin at 18 °C, and 10 µg/mL Cytochalasin D was added after 179 the crushing spin. Extracts with intact F-actin also demixed at similar crowding factors. 180 181

Crowding of Xenopus egg extracts 182
Xenopus egg extracts were crowded by two methods, using Sephadex G-25 resin or 30 kDa MWCO filter 183 units, which gave similar results. To crowd using coarse Sephadex G-25 gel filtration resin (Sigma-Aldrich 184 Cat#GE17-0034-01), 30 µg dry resin was added to 150 µL extracts in a PCR tube. The resin was submerged 185 and dispersed with a pipette tip, then the slurry was incubated for 5 min on ice. Then several holes were 186 punched in the bottom of the PCR tube using the tip of a 27G needle, which makes holes small enough to 187 retain resin in the tube. Then the PCR tube was placed inside a 0.5 mL tube, which was in turn placed 188 inside a 1.5 mL tube for centrifugation. The tubes were centrifuged at 4000 rcf for 4 min to collect the 189 crowded extract in the 0.5 mL tubes. To crowd using filter units, extracts were centrifuged in Amicon filter

Measurement of protein concentration 214
Protein concentration in each phase was measured by Micro BCA following TCA precipitation. 215 216

Isolation of phases by centrifugation 217
Crowded extracts were centrifuged at 20000 rcf for 20 min. After centrifugation, the less dense R phase 218 was aspirated into an 18 G blunt needle, carefully as not to disturb the interface between the R and G 219 phases. The R phase appeared as two opaque layers of slightly different colors and both these were 220 included in the R sample. Then a hole was punched in the bottom of the tube, and by pressing the top of 221 the tube, the higher density G phase was pushed through the hole, likewise carefully as to avoid the 222 interface between the phases. 223 224

Electron microscopy 225
Extract was spread on coverslips then samples were prepared by standard methods. Extract samples were 226 fixed with 1.5% glutaraldehyde in 0.1 M cacodylate buffer pH 7.4, post fixed with 1% osmium 227 tetroxide/potassium ferrocyanide, en block stained with 1% uranyl acetate, dehydrated and embedded in 228 Epon Araldite, then sectioned and on grid stained with uranyl acetate and lead citrate. Samples were 229 viewed on a Tecnai G-2 BioTwin electron microscope and imaged with an AMT CCD camera. 230 231

Mass spectrometry 232
Samples were denatured in 5 M guanidine thiocyanate, 5 mM dithiothreitol (DTT) (US Biological #D8070) 233 for 10 min at 60 °C, then cysteines were alkylated with N-ethylmaleimide (NEM). The eluate was 234 precipitated with trichloroacetic acid then subjected to proteolysis followed by the MultiNotch MS3 235 method as described McAlister et al., 2014;Sonnett et al., 2018), with channels 236 normalized by the total number of counts. Native MWs were based on previous estimates (Wühr et al., 237 2015). Gene Ontology terms used for Fig   Video S1. Liquid behavior of phases. (Related to Fig 1D) 1.4x crowded extract was confined between coverslips then imaged immediately to observe spreading flow. Both phases exhibited liquid behavior, including deformation under flow, splitting, fusion, and rounding by surface tension. Extract was imaged by differential interference contrast (DIC) microscopy with a 20x objective.